Rayleigh Ritz Method Method for Discrete Systems

Structural Dynamics: 2CE 611

Dr S S Mishra, NIT Patna

Let us have a recap of the facts of Rayleigh Quotient:

We know that Rayleigh quotient provides an upper bound for the lowest frequency λ1, that is

RQ ≥ λ1

The above inequality can also be interpreted as

λ1= ω21=min RQ(U) = RQ(U1)

The meaning is that :

(1) Rayleigh quotient has a minimum value at the exact fundamental modal vector. The
corresponding minimum value is the fundamental eigen value.
(2) Eq(1) also say that regardless of the value of the trial vector u the Rayleigh quotient is
always larger than the lowest eigen value.
(3) The implication is that the Rayleigh Quotient has a stationary value at an actual
eigenvector φr where the stationary value is the associated eigenvalue λr ( r =1,
2…n).

Rayleigh method is a powerful tool to estimate the systems natural frequencies.

Ritz extended the Rayleigh method to generalize the concept of an estimated vector. Rather than
using a trial single vector as an estimated vector he used several trial vector to get the estimated
vector. First he assumed that each trial vector is an eigen vector.

Let the trial vectors are xˆ 1 , xˆ 2, .......

If they were really the eigen vectors , the Rayleigh quotient will be stationary with respect to the
perturbation around them. He thought that if the independent trial vectors are combined together
they will best approximate the eigen vectors.

Then the estimated eigen vector (let us say E )

 n

E= ∑ b x̂
i =1
i i (1)

The unknown coefficients bi were determined by forcing the RQ (E) to be stationary.

The Rayleigh quotient in terms of estimated eigen vectors

E T kE
RQ ( E ) = (2)
E T mE

Since E is a function of constants b1, b2, b3, etc. , the Rayleigh quotient becomes a
function of the undetermined coefficients b1, b2, b3, … b3 alone.

RQ ( E) = RQ (b1, b2, b3, … bn) (3)

The stationary condition can be written as

∂RQ ∂RQ ∂RQ

δRQ = δb1 δb2 + ........ δbn = 0 (4)
∂b1 ∂b2 ∂bn

Since trial functions are independent these coefficients are are also independent which in turn
implies that the variations in the coefficients are also independent. Eq 4 is satisfied only when
the variations are all equal to zero. Then the necessary condition for stationarity is gien as

∂RQ
=0 (5)
∂bi

If we write the Rayleigh coefficient as

E T kE N (b1 , b2 ,...bn )
RQ ( E ) = = (6)
E T m E D(b1 , b2 ,...bn )

where, numerator N= ETkE and denominator D= ETmE then the necessary stationarity
condition

∂RQ D(∂N / ∂bi ) − N (∂D / ∂bi ) 1  ∂N N ∂D  1  ∂N ∂D 

= 2
=  − =  − RQ =0 (7)
∂bi D D  ∂bi D ∂bi  D  ∂bi ∂bi 

It can shown that after some manipulation that

n n

N= ∑∑ b b K
i =1 j =1
i j ij (8)
 n n

D= ∑∑ b b M
i =1 j =1
i j ij (9)

Where K ij = xˆ iT kxˆ j (10)

and M ij = xˆ iT mxˆ j (11)

For a typical i, the last term of Eq (7)

∂N
=2(b1Ki1 + b2Ki2 + b3Ki3 ….)
∂bi

which can be written as,

b1 
b 
∂N  
= 2[K i1 Ki2 K i3 .] 2 
∂bi b3 
 . 

Similarly,

b1 
b 
∂D  
= 2[M i1 M i2 M i3 .] 2 
∂bi b3 
 . 

Kij and Mij are the elements of K and M Matrices, then the eigen value problem with these K
and M becomes

 K 11 K12 .  b1   M 11 M 12 .   b1  0

K      
 21 K 22 .  b2  − RQ  M 21 M 22 .  b2  = 0 (12)
 . . K nn  b3   . . M nn  b3  0

Or in compact form

[K]B-RQ[M]B=0 (13)
Example: We shall start with a simple example and see the results of Rayleigh –Ritz approach
to the actual values. The system is shown in Figure. The problem is use two trial vectors to
analyze the problem

x1 x2 x3
k1 k2 k3 k4
m1 m2 m3

Given : k1= k2=k3= k4=1 and m1=m2=m3=1

For this system the eigen pairs are ω12 = 0.5858(rad / s) 2 , ω 22 = 2.000(rad / s) 2 and
2 2
ω = 3.4142(rad / s )
31

1  1  1 
     
And exact eigen vectors are x1 =  2  , x 2 =  0  and x 3 = − 2 
1  − 1  1 
     

Solution: we are using two trial vectors. This means we are treating this as two degree of
freedom system.

4  0
   
Let us use two trial vectors xˆ 1 = 3 and xˆ 2 = 2
0 4 
   

Let us form K and M matrices from Eq (10 and (11)

T
 4 0  2 − 1 0  4 0 
 26 − 8
K = 3 2 − 1 2 − 1 3 2 = 
− 8 24 
0 4  0 − 1 2  0 4 

T
4 0 1 0 0 4 0
25 6 
And M = 3 2 0 1 0 3 2 = 
 6 20
0 4 0 0 1  0 4

The eigen value problem using Eq12 or Eq 13 becomes

  26 − 8 b1  25 6  b1 
− 8 24  b  − RQ  6 20 b  =0
  2    2 

Solving this 2x2 eigen value problem

0.2500
We get RQ1 =0.5961 (rad/s)2 and B1 =  
0.2396

 0.2500 
RQ2 =2.0246 (rad/s)2, B2 =  
− 0.3054

A comparison of the estimated eigen values with the exact eigen values gives an
exciting result!

Now let us find out the estimated eigen vector from Eq (1)

4  0 1.000 
     
E1 = (B1 (1,1) )xˆ 1 + (B1 (2,1) ) xˆ 2 = 0.25003 + 0.23962 = 1.229 
0 4 0.959
     

4 0  1.000 

     
E 2 = (B2 (1,1) ) xˆ 1 + (B2 (2,1) ) xˆ 2 = 0.25003 − 0.30542 =  0.1392 
0 4 − 1.2216
     

We see that less error is associated with the eigen values than with the eigen vector how
ever they are some what close to the actual eigen vector. The assumed trial vectors are
arbitrary.

EXERCISE: Now for the same problem let us add one more vector and see whether
improvement in results takes place. Let us consider now

4  0 1
    
trial vectors xˆ 1 = 3 xˆ 2 = 2 xˆ 2 = 1
0 4  1
    

Determine now the Rayleigh Quotient (eigen values) and the estimated eigen vectors. Comment
on the results.

NOTE: WHAT WE ARE DOING in the Rayleigh –Ritz method?

We are forcing the system to vibrate as per our assumed vector!

Rayleigh- Ritz Method for Continuous Systems

In the continuous system we have infinite degrees- of- freedom. The trial functions are chosen as
a continuous function. A set of trial functions are chosen to estimate the eigen function or
displacement functions.

Let our trial functions are denoted as ψi(x), then

n

y ( x) = ∑ b ψ ( x)
1
i i (1)

Here ψ i (x) are the independent trial functions and bi are the undermined coefficients. Our
ojective is to produce n degree of freedom discrete model under the assumption of the trial
functions. The coefficients bi are to be so adjusted to render the Rayleigh Quotient stationary.
Obvviosly the weak form of the eigen value problemhas to be solved to determine the
coefficients

Proceeding as per the discrets system, the

Vmax
The Rayleigh Quotient = RQ(y) RQ ( y ) = λ = ω 2 = (2)
Tref

n n

∑∑ K b b
1
Where Vmax = ij i j (3)
2 i =1 j =1

Kij depends on the type of problem and the trial function , for example for an axial rod
L

∫
And K ij = K ji = EA( x)ψ i' ( x)ψ 'j ( x)dx
0
(4)

n n

∑∑ M b b
1
Tref = ij i j (5)
2 i =1 j =1
 L

∫
Where, M ij = M ji = m( x)ψ i ( x)ψ j ( x)dx
0
(6)